app.py

app.py

@@ -3,158 +3,234 @@ import gradio as gr
 import wikipedia
 import numpy as np
 import faiss
-from langdetect import detect
 from gtts import gTTS
-from transformers import pipeline
-from sentence_transformers import SentenceTransformer
 import tempfile
+from langdetect import detect
 import speech_recognition as sr
 from pydub import AudioSegment
-from functools import lru_cache
-
-# --- Load models ---
-models = {}
-def load_models():
-    models['encoder'] = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
-    models['to_en'] = pipeline('translation', model='Helsinki-NLP/opus-mt-mul-en')
-    for lang in ['fr', 'ar', 'zh', 'es']:
-        models[f'en_to_{lang}'] = pipeline('translation_en_to_' + lang, model=f'Helsinki-NLP/opus-mt-en-{lang}')
-    models['answer_gen'] = pipeline('text2text-generation', model='google/flan-t5-base', max_length=1024)
-load_models()
-
-# --- Utility functions ---
-def detect_language(text):
-    try:
-        return detect(text)
-    except:
-        return 'en'
+from transformers import pipeline # SentenceTransformer is not here
+from sentence_transformers import SentenceTransformer # Correct import
+import os
+from pydub.silence import split_on_silence
+import time
+import pytesseract
+
+
+# Initialize models
+models = {
+    'translator': pipeline('translation', model='Helsinki-NLP/opus-mt-mul-en'),
+    'answer_gen': pipeline('text2text-generation', model='google/flan-t5-base'),
+    'encoder': SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
+}
+
+# Add translation models
+for lang in ['fr', 'ar', 'zh', 'es']:
+    models[f'en_to_{lang}'] = pipeline(f'translation_en_to_{lang}',
+                                     model=f'Helsinki-NLP/opus-mt-en-{lang}')
 
 def translate(text, src, tgt):
-    if src == tgt:
-        return text
-    if src != 'en':
-        text = models['to_en'](text)[0]['translation_text']
-    if f'en_to_{tgt}' in models:
-        return models[f'en_to_{tgt}'](text)[0]['translation_text']
+    if src == tgt: return text
+    if src != 'en': text = models['translator'](text)[0]['translation_text']
+    if f'en_to_{tgt}' in models: return models[f'en_to_{tgt}'](text)[0]['translation_text']
     return text
 
-def tts_play(text, lang):
-    tts = gTTS(text=text, lang=lang)
-    path = tempfile.mktemp(suffix='.mp3')
-    tts.save(path)
-    return path
-
-def chunk_text(text, max_words=100):
-    sentences = text.split('. ')
-    chunks, current, length = [], [], 0
-    for sent in sentences:
-        words = sent.split()
-        if length + len(words) > max_words:
-            chunks.append(' '.join(current))
-            current, length = [sent], len(words)
-        else:
-            current.append(sent)
-            length += len(words)
-    if current: chunks.append(' '.join(current))
-    return chunks
-
-def build_faiss_index(chunks, model):
-    emb = model.encode(chunks, convert_to_numpy=True)
-    index = faiss.IndexFlatL2(emb.shape[1])
-    index.add(emb)
-    return index
-
-@lru_cache(maxsize=20)
-def prepare_faiss_for_topic(topic):
-    wikipedia.set_lang('en')
-    page = wikipedia.page(topic)
-    chunks = chunk_text(page.content)  # Use full content, no slicing limit
-    return chunks, build_faiss_index(chunks, models['encoder'])
-
-def retrieve_context(q, idx, chunks, model, top_k=5):
-    emb = model.encode([q], convert_to_numpy=True)
-    _, inds = idx.search(emb, top_k)
-    return ' '.join(chunks[i] for i in inds[0])
-
-# --- Main Q&A function ---
-def qa_system(audio, text_q, topic, lang, history):
-    q = ''
-    if audio:
-        try:
-            r = sr.Recognizer()
-            wav = tempfile.mktemp('.wav')
-            AudioSegment.from_file(audio).export(wav, format='wav')
-            with sr.AudioFile(wav) as src:
-                q = r.recognize_google(r.record(src))
-        except Exception as e:
-            return f"❌ Could not transcribe audio: {e}", None, history, ''
-    elif text_q:
-        q = text_q.strip()
-    else:
-        return '❌ Please speak or type your question.', None, history, ''
-
+def text_to_speech(text, lang):
     try:
-        chunks, idx = prepare_faiss_for_topic(topic)
+        tts = gTTS(text=text, lang=lang)
+        audio_path = tempfile.mktemp(suffix='.mp3')
+        tts.save(audio_path)
+        return audio_path
     except Exception as e:
-        return f'Error loading content: {e}', None, history, ''
-
-    ctx = retrieve_context(q, idx, chunks, models['encoder'])
-    q_en = translate(q, detect_language(q), 'en')
-
-    # Debug prints — remove in production
-    print("Question (original):", q)
-    print("Question (English):", q_en)
-    print("Retrieved context snippet:", ctx[:500], "...\n")
-
-    prompt = f"Context:\n{ctx}\n\nQuestion: {q_en}\nAnswer:"
-    ans_en = models['answer_gen'](prompt)[0]['generated_text']
+        print(f"TTS Error: {e}")
+        return None
+
+def process_audio(audio_path):
+    recognizer = sr.Recognizer()
+    sound = AudioSegment.from_file(audio_path)
+    chunks = split_on_silence(sound,
+        min_silence_len=500,
+        silence_thresh=sound.dBFS-14,
+        keep_silence=500
+    )
+
+    full_text = ""
+    for chunk in chunks:
+        chunk_path = tempfile.mktemp(suffix='.wav')
+        chunk.export(chunk_path, format="wav")
+        with sr.AudioFile(chunk_path) as source:
+            audio = recognizer.record(source)
+            try:
+                text = recognizer.recognize_google(audio)
+                full_text += f" {text}"
+            except:
+                continue
+        os.unlink(chunk_path)
+
+    return full_text.strip() if full_text else None
+
+def get_wikipedia_content(topic):
+    try:
+        wikipedia.set_lang('en')
+        try:
+            page = wikipedia.page(topic, auto_suggest=False)
+            return page.summary[:1000]
+        except wikipedia.exceptions.DisambiguationError as e:
+            page = wikipedia.page(e.options[0])
+            return page.summary[:1000]
+    except Exception as e:
+        print(f"Wikipedia error: {e}")
+        return None
+
+def generate_response(text, topic, lang):
+    context = get_wikipedia_content(topic)
+    if not context:
+        return "Could not find information. Please try another topic.", None
+
+    prompt = f"Context: {context}\nQuestion: {text}\nAnswer:"
+    answer = models['answer_gen'](prompt, max_length=200)[0]['generated_text']
+    translated = translate(answer, 'en', lang) if lang != 'en' else answer
+    audio_path = text_to_speech(translated, lang)
+
+    return translated, audio_path
+
+def handle_interaction(audio, text, topic, lang, chat_history):
+    if audio is not None:
+        recognized_text = process_audio(audio)
+        if recognized_text:
+            text = recognized_text
+        else:
+            chat_history.append(("", "Could not understand audio. Please try again."))
+            return chat_history, "", None
 
-    print("Generated answer (English):", ans_en)
+    if not text.strip():
+        chat_history.append(("", "Please enter a question."))
+        return chat_history, "", None
 
-    ans = ans_en if lang == 'en' else translate(ans_en, 'en', lang)
-    audio_path = tts_play(ans, lang)
-    history.append((q, ans))
-    chat = '\n\n'.join(f"Q{i+1}: {x}\nA{i+1}: {y}" for i,(x,y) in enumerate(history))
-    return f'You asked: {q}\n\nAnswer: {ans}', audio_path, history, chat
+    response, audio_output = generate_response(text, topic, lang)
+    chat_history.append((text, response))
 
-def clear_all():
-    return None, '', None, [], ''
+    return chat_history, "", audio_output
 
-# --- Gradio UI with styling ---
-css_style = """
+# Custom CSS with light blue and dark blue theme
+custom_css = """
 .gradio-container {
-    background-color: #cce7ff !important;  /* Light blue */
-    border: 3px solid #000022 !important; /* Balanced blue-black border */
-    border-radius: 12px;
+    background: #f0f8ff !important;
+    border: 3px solid #00008b !important;
+    border-radius: 10px !important;
+    font-family: 'Arial', sans-serif;
+}
+.gr-box {
+    background-color: #e6f2ff !important;
+    border: 2px solid #00008b !important;
+    border-radius: 8px !important;
+}
+.gr-button {
+    background-color: #4d94ff !important;
+    border: 2px solid #00008b !important;
+    color: white !important;
+    border-radius: 6px !important;
+}
+.gr-button:hover {
+    background-color: #1a75ff !important;
+}
+.gr-chatbot {
+    background-color: #e6f2ff !important;
+    border: 2px solid #00008b !important;
+    border-radius: 8px !important;
+}
+.gr-textbox, .gr-dropdown, .gr-audio {
+    background-color: #e6f2ff !important;
+    border: 2px solid #00008b !important;
+    border-radius: 6px !important;
+}
+.welcome-header {
+    text-align: center;
+    color: #00008b !important;
+    margin-bottom: 20px;
+}
+.welcome-message {
+    background-color: #e6f2ff;
     padding: 20px;
+    border-radius: 10px;
+    border: 2px solid #00008b;
+    margin-bottom: 20px;
+}
+.avatar {
+    width: 80px;
+    height: 80px;
+    margin: 0 auto;
+    display: block;
 }
 """
 
-with gr.Blocks(css=css_style) as demo:
-    gr.Markdown("""
-    <h1 style='color:#003366; text-align:center; margin-bottom: 0;'>🌐 Multilingual Wikipedia Q&A Assistant</h1>
-    <p style='text-align:center; font-size:16px; margin-top: 0;'>Ask your questions by typing or speaking, and get answers in your language!</p>
-    """)
-
-    state = gr.State([])
-
-    with gr.Row():
-        ai = gr.Audio(type='filepath', label='🎤 Speak your question')
-        ti = gr.Textbox(lines=3, placeholder='Or type your question here')
+# Welcome page content
+welcome_html = """
+<div class="welcome-header">
+    <img src="https://i.imgur.com/6wBs5mO.png" class="avatar" alt="AI Assistant">
+    <h1>Welcome to Your Multilingual AI Assistant! 🌍</h1>
+</div>
+<div class="welcome-message">
+    <h3>Hello! I'm your personal Wikipedia assistant 🤖</h3>
+    <p>I can help you find information on any topic in multiple languages. Here's what I can do:</p>
+    <ul>
+        <li>🔍 Answer questions from Wikipedia knowledge</li>
+        <li>🗣️ Understand both voice and text input</li>
+        <li>🌐 Respond in English, French, Spanish, Chinese, or Arabic</li>
+        <li>🔊 Speak answers back to you</li>
+    </ul>
+    <p>To get started, simply type your question or click the microphone to speak!</p>
+</div>
+"""
 
-    with gr.Row():
-        tp = gr.Textbox(value='Artificial intelligence', label='Wikipedia Topic')
-        lg = gr.Dropdown(['en','am','fr','ar','es','zh'], value='en', label='Output Language')
+with gr.Blocks(css=custom_css, title="🌍 Multilingual AI Assistant") as demo:
+    # Welcome page
+    gr.HTML(welcome_html)
 
+    # Main interface
     with gr.Row():
-        sb = gr.Button('🔍 Get Answer')
-        cb = gr.Button('🗑️ Clear All')
-
-    ao = gr.Textbox(label='🤖 Answer')
-    av = gr.Audio(label='🔊 Listen Answer')
-    cd = gr.Markdown(label='🗂️ Chat History')
-
-    sb.click(qa_system, inputs=[ai, ti, tp, lg, state], outputs=[ao, av, state, cd])
-    cb.click(clear_all, outputs=[ai, ti, tp, state, cd])
-
-demo.launch(share=True)
+        with gr.Column(scale=1):
+            audio_input = gr.Audio(
+                sources=["microphone", "upload"],
+                type="filepath",
+                label="🎤 Speak or upload audio",
+                interactive=True
+            )
+            topic_input = gr.Textbox(
+                "Artificial Intelligence",
+                label="📚 Wikipedia Topic"
+            )
+            lang_input = gr.Dropdown(
+                ["en", "fr", "es", "zh", "ar"],
+                value="en",
+                label="🌐 Output Language"
+            )
+
+        with gr.Column(scale=2):
+            chatbot = gr.Chatbot(label="Conversation")
+            text_input = gr.Textbox(
+                placeholder="Type your question here...",
+                label="✏️ Or type here"
+            )
+            with gr.Row():
+                clear_btn = gr.Button("🗑️ Clear Chat")
+                submit_btn = gr.Button("🚀 Submit", variant="primary")
+
+    audio_output = gr.Audio(label="🔊 Answer", visible=True)
+
+    # Event handlers
+    submit_btn.click(
+        handle_interaction,
+        inputs=[audio_input, text_input, topic_input, lang_input, chatbot],
+        outputs=[chatbot, text_input, audio_output]
+    )
+    text_input.submit(
+        handle_interaction,
+        inputs=[audio_input, text_input, topic_input, lang_input, chatbot],
+        outputs=[chatbot, text_input, audio_output]
+    )
+    clear_btn.click(
+        lambda: ([], "", None),
+        outputs=[chatbot, text_input, audio_output]
+    )
+
+demo.launch(share=True)